VGMGC: Variational Graph Generator for Multi-View Graph Clustering

This is the code of paper: Variational Graph Generator for Multi-View Graph Clustering.

We sincerely appreciate it if you cite this paper as

@ARTICLE{10833915,
  author={Chen, Jianpeng and Ling, Yawen and Xu, Jie and Ren, Yazhou and Huang, Shudong and Pu, Xiaorong and Hao, Zhifeng and Yu, Philip S. and He, Lifang},
  journal={IEEE Transactions on Neural Networks and Learning Systems}, 
  title={Variational Graph Generator for Multiview Graph Clustering}, 
  year={2025},
  volume={},
  number={},
  pages={1-14},
  doi={10.1109/TNNLS.2024.3524205}}

Requirements

• Python >= 3.8
• Pytorch >= 1.11.0
• munkres >= 1.1.4
• scikit-learn >= 1.0.1
• scipy >= 1.8.0

Datasets

ACM and DBLP are included in ./data/. The other datasets are public available.

Dataset	#Clusters	#Nodes	#Features	Graphs
ACM	3	3025	1830	$\mathcal{G}^1$ co-paper $\mathcal{G}^2$ co-subject
DBLP	4	4057	334	$\mathcal{G}^1$ co-author $\mathcal{G}^2$ co-conference $\mathcal{G}^3$ co-term
Amazon photos	8	7487	745 7487	$\mathcal{G}^1$ co-purchase
Amazon computers	10	13381	767 13381	$\mathcal{G}^1$ co-purchase

Test VGMGC

# Test VGMGC on ACM dataset
python vgmgc.py --dataset acm --train False --model_name vgmgc_acm.pkl --order 8 --lam_emd 1

# Test VGMGC on DBLP dataset
python vgmgc.py --dataset dblp --train False --model_name vgmgc_dblp.pkl --order 8 --lam_emd 5

# Test VGMGC on Cora dataset
python vgmgc.py --dataset cora --train False --model_name vgmgc_cora.pkl --order 10 --weight_soft 1. --min_belief 0.2 --max_belief 0.99 --lam_emd 0.2 --kl_step 5 --lam_elbo_kl 1 --threshold 0.5 --temperature 1 --add_graph True 

# Test VGMGC on Citeseer dataset
python vgmgc.py --dataset citeseer --train False --model_name vgmgc_citeseer.pkl --order 8 --weight_soft 1. --min_belief 0.2 --max_belief 0.99 --lam_emd 1. --kl_step 5 --lam_elbo_kl 1 --threshold 0.5 --temperature 1 --add_graph True

# Test VGMGC on 3Sources dataset
python vgmgc.py --dataset 3sources --train False --model_name vgmgc_3sources_acc0.9467.pkl --order 1 --weight_soft 1. --min_belief 0.2 --max_belief 0.99 --lam_emd 10. --kl_step 5 --lam_elbo_kl 1 --threshold 0.5 --temperature 1

# Test VGMGC on bbc sport  dataset
python vgmgc.py --dataset bbcsport_2view --train False --model_name vgmgc_bbcsport_2view_acc0.9835.pkl --order 2 --weight_soft 1. --min_belief 0.2 --max_belief 0.99 --lam_emd 100. --kl_step 5 --lam_elbo_kl 1 --threshold 0.5 --temperature 1

Train VGMGC

# Train VGMGC on ACM dataset
python vgmgc.py --dataset acm --train true --model_name vgmgc_acm1.pkl --order 8 --weight_soft 0.9 --min_belief 0.7 --max_belief 0.99 --lam_emd 1 --kl_step 5 --lam_elbo_kl 1 --threshold 0.8 --temperature 5

# Train VGMGC on DBLP dataset
python vgmgc.py --dataset dblp --train true --model_name vgmgc_dblp1.pkl --order 8 --weight_soft 0.1 --min_belief 0.2 --max_belief 0.99 --lam_emd 5 --kl_step 10 --lam_elbo_kl 1 --threshold 0.8 --temperature 1


python vgmgc.py --dataset cora --train true --model_name vgmgc_cora1.pkl --order 10 --weight_soft 1. --min_belief 0.2 --max_belief 0.99 --lam_emd 0.2 --kl_step 5 --lam_elbo_kl 1 --threshold 0.5 --temperature 1  --add_graph True --update_interval 2


python vgmgc.py --dataset citeseer --train true --model_name vgmgc_citeseer1.pkl --order 8 --weight_soft 1. --min_belief 0.2 --max_belief 0.99 --lam_emd 1. --kl_step 5 --lam_elbo_kl 1 --threshold 0.5 --temperature 1  --add_graph True --update_interval 2


python vgmgc.py --dataset 3sources --train true  --order 1 --weight_soft 1. --min_belief 0.2 --max_belief 0.99 --lam_emd 10. --kl_step 5 --lam_elbo_kl 1 --threshold 0.5 --temperature 1 --latent_dim 512 --hidden_dim 512 --update_interval 2


python vgmgc.py --dataset bbcsport_2view --train true  --order 2 --weight_soft 1. --min_belief 0.2 --max_belief 0.99 --lam_emd 100. --kl_step 5 --lam_elbo_kl 1 --threshold 0.5 --temperature 1 --latent_dim 512 --hidden_dim 512 --update_interval 2

Parameters: More parameters and descriptions can be found in the script and paper.

Results of VGMGC

	NMI%	ARI%	ACC%	F1%
ACM	77.3	83.7	94.3	94.3
DBLP	78.3	83.7	93.2	92.7
Amazon photos	66.8	58.4	78.5	76.9
Amazon computers	53.5	47.5	62.2	50.2

	NMI%	ARI%	ACC%	F1%
Cora	55.6	51.7	73.5	71.6
Citeseer	45.2	46.4	70.1	65.4
BBC sport	94.6	95.1	98.3	98.6
3sources	86.5	88.1	94.7	93.5